It's easy to think that GenAI is always the best AI. Not true! It is one of several tools, each playing a distinct role. And these tools help each other.
The report "Adopting Generative AI in Banking" from Deutsche Bank and Kodex AI, distinguishes robotic process automation (RPA), traditional AI, generative AI and agentic AI.
Below are some excerpts.
Kudos to authors Thomas Kaiser, Boon-Hiong Chan and Delane Zahoruiko for their in-depth work. And thanks to Tobias Zwingmann and Donald Farmer for flagging the report on LinkedIn.
The report is an educational view of the bleeding edge from an early adopter. Would welcome your thoughts.
"As GenAI and the range of AI technologies continue to evolve, their abilities to solve complex situational challenges would also expand and appear to be infinite.
"As expectation builds with each news of new capabilities or of another successful use case, GenAI systems risk becoming the silver bullet to everything that needs to be solved, which is unrealistic.
"Hence, in deploying GenAI, which is a highly powerful tool that augments and creates super productivity benefits when properly applied, it is important to understand what and where it should be deployed with grounded expectations.
"This would facilitate business case success and fit-for-purpose governance.
"This paper examines the GenAI system, with Figure 2 highlighting key differences between GenAI, AI, and other comparable systems. While these systems may seem alike, each possesses unique capabilities, risks, and regulatory profiles...
> RPA
"Automates repetitive tasks and workflow. No “new” output.
"Copies human interactions with systems. Does not create new methods of interaction"
> Traditional AI
"Pattern recognition, regression analysis/prediction, classification.
"Analysis, application and prediction based on existing data/ model. Arguably little real time learning.
> GenAI
"Content generation (eg, text, images, code, etc).
"Outputs new data and generate output. Real-time learning, self-course correction.
> Agentic AI
"Autonomous decision making and action.
"Interacts with other systems, learn and act in real time.
My take:
Successful adopters will develop applications and workflows that apply two or more of these techniques to their own governed domain-specific data.
They will roll them out to expert internal users that can inspect, oversee and correct the outputs before putting anything into action.
They will start with small, low-risk projects that make tactical changes to reduce existing business pain.
If this works, they will expand their AI initiative to tackle bigger challenges.
#ai #genai #rpa #data #innovation
Thomas Zeutschler Gernot Molin Frank Casale Luda Kopeikina, PhD in AI, MIT Sloan Fellow EM360Tech
